制药企业和患者数据

Price Waterhouse Coopers predicts that sharing personal health information beyond the direct care of the patient will be a two billion dollar market over the next few years [1].
Many companies thrive through selling data based on acquiring, curating and aggregating personal data. For example, IMS Health collects personal prescription information from pharmacies and pharmacy benefits programs, and then uses it to sell market information to pharmaceutical companies [2]. Acxiom collects personal information from public records, such as marriage licenses and voter lists, and uses it to provide background checks [3]. Geisinger Health System, a large integrated health system, created a company called MedMining, which licenses its data to promote healthcare research, primarily to major pharmaceutical companies and large biotech companies [4].
1 PriceWaterhouseCoopers. Transforming healthcare through secondary use of health data. 2009. 2 IMS Health. IMS Facts at a Glance. As of September 30, 2010, http://www.imshealth.com/ 3 Acxiom. FAQs and EEOC Guidelines. As of Septemer 30, 2010 http://www.acxiom.com/products_and_services/background_screening/faq/Pages/FAQs.aspx 4 MedMining. Welcome to MedMining. As of September 30, 2010 http://www.medmining.com/

https://www.mckinsey.com/industries/pharmaceuticals-and-medical-products/our-insights/how-big-data-can-revolutionize-pharmaceutical-r-and-d

市场营销

One role of pharmaceutical research companies is to provide information about their medicines to health care professionals. T his interaction between pharmaceutical representatives and health care professionals is often referred to as “marketing and promotion.” Without it, health care professionals would be less likely to have the latest, accurate information available regarding pre - scription medicines, which play an increasing role in effective health care phrma_marketing_brochure_influences_on_prescribing_final.pdf

线上推广自己官网收集患者信息推荐药品 –Reimbursement Programs –Disease management programs –Pharmacy benefits programs

R&D

–DNA ? –Clinical trials ? –Drug safety monitoring –Biostatistical analysis

clinical-trial efficiency A combination of new, smarter devices and fluid data exchange will enable improvements in clinical-trial design and outcomes as well as greater efficiency. Clinical trials will become increasingly adaptable to react to drug-safety signals seen only in small but identifiable subpopulations of patients. Examples of potential clinical-trial efficiency gains include the following:

Dynamic sample-size estimation (or reestimation) and other protocol changes could enable rapid responses to emerging insights from the clinical data. Efficiency gains are achieved by enabling smaller trials for equivalent power or shortening the time necessary to expand a trial.
Adapting to differences in site patient-recruitment rates would allow a pharmaceutical company to address lagging sites, bring new sites online if necessary, and increase recruiting from more successful sites.
Increased use of electronic data capture could help in recording patient information in the provider’s electronic medical records. Using electronic medical records as the primary source for clinical-trial data rather than having a separate system could accelerate trials and reduce the likelihood of data errors caused by manual or duplicate entry.
Next-generation remote monitoring of sites, enabled by fluid, real-time data access, could improve management and responses to issues that arise in trials.

analyze drug efficacy,

Improve safety and risk management

Pharmaceutical companies can use safety as a competitive advantage in regulatory submissions and after regulatory approval, once the drug is on the market. Safety monitoring is moving beyond traditional approaches to sophisticated methods that identify possible safety signals arising from rare adverse events. Furthermore, signals could be detected from a range of sources, for example, patient inquiries on Web sites and search engines. Online physician communities, electronic medical records, and consumer-generated media are also potential sources of early signals regarding safety issues and can provide data on the reach and reputation of different medicines. Bayesian analytical methods, which can identify adverse events from incoming data, could highlight rare or ambiguous safety signals with greater accuracy and speed.

An early response to physician and patient sentiments could prevent regulatory and public-relations backlashes. The FDA is investing in the evaluation of electronic health records through the Sentinel Initiative, a legally mandated electronic-surveillance system that links and analyzes health-care data from multiple sources. As part of this system, the FDA now has secured access to data concerning more than 120 million patients nationwide.

Outcomes or economics studies

create new economic models that combine the provision of drugs and services.

Sharpen focus on real-world evidence

Real-world outcomes are becoming more important to pharmaceutical companies as payors increasingly impose value-based pricing. These companies should respond to this cost-benefit pressure by pursuing drugs for which they can show differentiation through real-world outcomes, such as therapies targeted at specific patient populations. In addition, the FDA and other government organizations have created incentives for research on health economics and outcomes.

To expand their data beyond clinical trials, some leading pharmaceutical companies are creating proprietary data networks to gather, analyze, share, and respond to real-world outcomes and claims data. Partnerships with payors, providers, and other institutions are critical to these efforts.

enhance future drug sales, and

https://dataprivacylab.org/projects/identifiability/pharma1.pdf

市场营销

The pharmaceutical company had contracts with a number of managed care organizations in which the managed care organizations received rebates that depended on their use of the pharmaceutical company ʼ s products. These contracts required the managed care organizations to submit copies of patient prescription claims data after each quarter to allow the pharmaceutical company to validate and ultimately pay rebates as established by the contract.

Prescription number
Pharmacy ID
Date of fill
NDC number
Quantity
Plan/Prescription level
Plan ID
Plan Name
Pharmacy ZIP
Unit of measure
Dosage form
Diagnosis code
Days supply
Prescription type
Total number of prescriptions
Therapeutic class
Reimbursement date
New/refill code
Product description

pharma1.pdf

IMS health

国际销售用过MIDAS，不清楚数据采集过程。美国的销量，NPA (National Prescription Audit) 和Xponent是sell-out的数据，就是指多少药被病人接收到，采集方法是在药房/Long term care收处方数据。 NSP (National Sales Perspectives) 和DDD (Drug Distribution Data) 是sell-in数据，指的是多少药从制药公司“流出”，数据采集发生在IMS和制药商（医药公司）之间。此外IMS还有病人数据，从药房或保险公司收集处方。这些数据库的处方覆盖率不等。比如APLD (Anonymous Patient Level Data) 覆盖全国80%的处方（因为只算零售渠道所以大概有不到两亿人口吧），但是作为纵向数据库的Pharmetrics Plus就只有大概几千万人。不过应该也不会用病人数据来看“销量”。

https://www.scientificamerican.com/article/how-data-brokers-make-money-off-your-medical-records/

For decades researchers have run longitudinal studies to gain new insights into health and illness. By regularly recording information about the same individuals' medical history and care over many years, they have, for example, shown that lead from peeling paint damages children's brains and bodies and have demonstrated that high blood pressure and cholesterol levels contribute to heart disease and stroke. To this day, some of the original (and now at least 95-year-old) participants in the famous Framingham Heart Study, which began in 1948, still provide health information to study investigators.

Health researchers are not the only ones, however, who collect and analyze medical data over long periods. A growing number of companies specialize in gathering longitudinal information from hundreds of millions of hospitals' and doctors' records, as well as from prescription and insurance claims and laboratory tests. Pooling all these data turns them into a valuable commodity. Other businesses are willing to pay for the insights that they can glean from such collections to guide their investments in the pharmaceutical industry, for example, or more precisely tailor an advertising campaign promoting a new drug.

By law, the identities of everyone found in these commercial databases are supposed to be kept secret. Indeed, the organizations that sell medical information to data-mining companies strip their records of Social Security numbers, names and detailed addresses to protect people's privacy. But the data brokers also add unique numbers to the records they collect that allow them to match disparate pieces of information to the same individual—even if they do not know that person's name. This matching of information makes the overall collection more valuable, but as data-mining technology becomes ubiquitous, it also makes it easier to learn a previously anonymous individual's identity.

At present, the system is so opaque that many doctors, nurses and patients are unaware that the information they record or divulge in an electronic health record or the results from lab tests they request or consent to may be anonymized and sold. But they will not remain in the dark about these practices forever. In researching the medical-data-trading business for an upcoming book, I have found growing unease about the ever expanding sale of our medical information not just among privacy advocates but among health industry insiders as well.

The entire health care system depends on patients trusting that their information will be kept confidential. When they learn that others have insights into what happens between them and their medical providers, they may be less forthcoming in describing their conditions or in seeking help. More and more health care experts believe that it is time to adopt measures that give patients more control over their data. Multibillion-Dollar Business

The dominant player in the medical-data-trading industry is IMS Health, which recorded $2.6 billion in revenue in 2014. Founded in 1954, the company was taken private in 2010 and relaunched as public in 2014. Since then, it has proved an investor favorite, with shares rising more than 50 percent above its initial price in little more than a year. At press time, IMS was a $9-billion company. Competitors include Symphony Health Solutions and smaller rivals in various countries.

Decades ago, before computers came into widespread use, IMS field agents photographed thousands of prescription records at pharmacies for hundreds of clerks to transcribe—a slow and costly process. Nowadays IMS automatically receives petabytes (1015 bytes or more) of data from the computerized records held by pharmacies, insurance companies and other medical organizations—including federal and many state health departments. Three quarters of all retail pharmacies in the U.S. send some portion of their electronic records to IMS. All told, the company says it has assembled half a billion dossiers on individual patients from the U.S. to Australia.

IMS and other data brokers are not restricted by medical privacy rules in the U.S., because their records are designed to be anonymous—containing only year of birth, gender, partial zip code and doctor's name. The Health Insurance Portability and Accountability Act (HIPAA) of 1996, for instance, governs only the transfer of medical information that is tied directly to an individual's identity.

Even anonymized, the data command premium prices. Every year, for example, Pfizer spends $12 million to buy health data from a variety of sources, including IMS, according to Marc Berger, who oversees the analysis of anonymized patient data at Pfizer. But companies engaged in the data trade tend to keep the practice below the general public's radar.

Case in point: In the 1990s IMS started selling data on what individual U.S. physicians prescribe to patients to help drug companies tailor sales pitches to specific care providers. (HIPAA protects the identity of patients, not health care workers.) For years doctors did not realize that outsiders had insights on their prescribing habits. “At the time, it was taboo. It was forbidden to ever mention that topic,” says Shahram Ahari, who used such data as a pharmaceutical representative visiting doctors for Eli Lilly from 1999 to 2000 and is now completing a residency at the University of Rochester. “It was the big secret.” Asked for a response, an Eli Lilly spokesperson replied in an e-mail, “We have always been up front that we receive data from IMS.”

Eventually physicians caught on and complained. Some considered such data gathering a privacy invasion; others objected to commercial firms profiting from details about their practices. A few states passed laws banning the collection of physician-prescribing habits. IMS challenged those rules all the way to the U.S. Supreme Court and—despite the arguments of 36 states, the Department of Justice, and numerous medical and consumer-advocacy groups supporting data limits—won its case in 2011 on corporate “free speech” grounds. The practice continues to this day, much of the time beyond public notice. What Could Go Wrong?

Once upon a time, simply removing a person's name, address and Social Security number from a medical record may well have protected anonymity. Not so today. Straightforward data-mining tools can rummage through multiple databases containing anonymized and nonanonymized data to reidentify the individuals from their ostensibly private medical records.

Indeed, computer scientists have repeatedly shown how easy it can be to crack seemingly anonymous data sets. For example, Harvard University professor Latanya Sweeney used such methods when she was a graduate student at the Massachusetts Institute of Technology in 1997 to identify then Massachusetts governor William Weld in publicly available hospital records. All she had to do was compare the supposedly anonymous hospital data about state employees to voter registration rolls for the city of Cambridge, where she knew the governor lived. Soon she was able to zero in on certain records based on age and gender that could have only belonged to Weld and that detailed a recent visit he made to a hospital, including his diagnosis and the prescriptions he took home with him.

“It is getting easier and easier to identify people from anonymized data,” says Chesley Richards, director of the Office of Public Health Scientific Services at the Centers for Disease Control and Prevention. “You may not be identifiable from a particular data set that an entity has collected, but if you are a broker that is assembling a number of sets and looking for ways to link those data, that's where, potentially, the risk becomes greater for identification.”

IMS officials say they have no interest in identifying patients and take careful steps to preserve anonymity. Moreover, there are no publicly recorded instances of someone taking anonymized patient data from IMS or a rival company and reidentifying individuals. Yet IMS does not want to talk too much about the gathering and selling of longitudinal data. At IMS, the CEO, the head of its Institute for Healthcare Informatics, the vice president of industry relations and the chief privacy officer declined to be interviewed for this article, but a company spokesperson did assist with fact-checking. Where to Draw the Line?

Apart from making money selling information to other businesses, IMS also shares some data with academic and other researchers for free or at a discount. The company has published a long list of medical articles that relied on its longitudinal data. For example, researchers learned that newer cardiovascular drugs reduce the length of hospital stays but do not prolong lives. In contrast, newer chemotherapy drugs are probably responsible for some of the recent decline in death rates from cancer in France.

Such benefits demonstrate that amassing medical data from multiple sources can have societal benefits. There is, however, a difference, says Jerry Avorn, a professor of medicine at Harvard Medical School, between “conscious, responsible researchers who only want to learn about medications' good and bad effects in a university medical school setting versus somebody sitting in the backroom [of a superstore] trying to figure out how can they sell more of product X by invading someone's privacy.”

One small step toward reestablishing trust in the confidentiality of medical information is to give individuals the chance to forbid collection of their information for commercial use—an option the Framingham study now offers its participants, as does the state of Rhode Island in its sharing of anonymized insurance claims. “I personally believe that at the end of the day, individuals own their data,” says Pfizer's Berger. “If somebody is using [their] data, they should know.” And if the collection is “only for commercial purposes, I think patients should have the ability to opt out.”

Seeking more detailed consent cannot, by itself, stem the erosion of patient privacy, but it will raise awareness—without which no further action is possible. Trust in the medical system is too vital to be sacrificed to uncontrolled market forces.

This reporting project was funded by a Reporting Award at New York University's Arthur L. Carter Journalism Institute.

This article was originally published with the title "For Sale: Your Medical Records"

The Hidden Trade in Our Medical Data: Why We Should Worry https://www.scientificamerican.com/article/the-hidden-trade-in-our-medical-data-why-we-should-worry/

For-profit companies use our anonymized medical data in a huge secondary market. Advances in computing make it increasingly possible for outsiders to identify people from among the hundreds of millions of patients in dossiers, putting intimate secrets about our bodies and minds at risk

By Adam Tanner on January 11, 2017

Excerpted and adapted from Our Bodies, Our Data: How Companies Make Billions Selling Our Medical Records. Copyright © by Adam Tanner. With permission of the publisher, Beacon Press. All Rights Reserved.

Companies that have nothing to do with our medical treatment are allowed to buy and sell our health care data, provided they remove certain fields of information, including birth date, name and Social Security number. These guidelines, outlined in the U.S. HIPAA rules, have allowed a multi-billion-dollar trade in anonymized patient data to emerge in recent years, with data mining firms collecting dossiers on hundreds of millions of patients. A growing number of data scientists and health care experts say the same computing advances that allow the aggregation of millions of anonymized patient files into a dossiers also make it increasingly possible to re-identify those files—that is, to match identities to patients.

“It's very difficult to protect data from re-identification through most processes that are used to anonymize it,” said Dr. Jonathan Wald, a Harvard Medical School instructor and expert on health data at the non-profit group RTI International. “That is easy when it is a rare condition and there are a few other tidbits. It is getting easier and easier because of the amount of electronic publicly available data and the amount of analytic engines to turn through it.”

Management Science Associates in Pittsburgh is one of the companies that helps data miners aggregate anonymized patient dossiers. Jani Syed, the company’s technical group director, surprised me with his candor about the risks of re-identification.

“In the area of big data there are always problems with the privacy,” he said. “No matter what you do, no matter how much data obfuscation you are going to do, if you have enough data it is always possible to identify a particular person. It's not that hard to do.”

Another way outsiders may be able to identify anonymized files is by cross referencing them with other sensitive files that hackers and thieves have obtained in recent years. Unfortunately, identified details about you from medical files may already be in circulation on the Internet or in hacker circles. This possibility is something I know about personally as I am one of many millions whom medical insurers and providers have notified as a victim of such attacks. Between 2009 and 2015, the U.S. Department of Health and Human Services recorded more than 1,300 data breaches involving more than 500 people, accessing data on more than 135 million people.

To date, there is no publicly recorded incident of hackers getting into the anonymized individual patient dossiers held by data miners, nor reported instances of re-identification of anonymized medical records in the United States other than academic experiments Even if thieves did hack such anonymized records, they would face the additional complication of re-identifying the records. The reward for all that effort would be a potentially richer array of insights into a patient than from single-source files, as anonymized patient data may contain pharmacy, claims, doctor, and even lab information.

Experts identify a variety of possible motivations for an outsider to seek to re-identify medical files.

A rival at work who wants your job or simply does not like you may know when you took medical leave and other clues that could make it possible to find you in batch of anonymous patient files. Suddenly, your re-identified files might appear in circulation. In a crime of passion, a romantic rival – or crossed former lover – might want to spread such information on the Internet, a variant of revenge porn in which former partners post intimate photos online.

“Health information, in particular, which can encompass a variety of things from sleep patterns to diagnoses to genetic markers, the data gathered about us can paint a very detailed and personal picture that is essentially impossible to de-identify, making it valuable for a variety of entities such as data brokers, marketers, law enforcement agencies, and criminals,” says Michelle De Mooy, director of the Privacy & Data Project at the Center for Democracy & Technology.

“Traditional methods of anonymization from commercial entities, such as the use of patient identifiers, have also become more of a problem with the amount of data available about individuals - there is of course an entire industry in vendors matching records retroactively.”

Medical data, both de-identified and re-identified, could also become national security weapons against members of the armed forces and their families, or high-ranking officers.

“It is not just that the information might embarrass a general or embarrass a senator—because we also see VIPs and so forth in our system—it is that the aggregation of certain health data in our context is potentially classified information,” said one military official who did not want to be named. “If I were to aggregate immunization data for a particular region of our country, like say Ft. Bragg, I might be able to learn where special operators are ready to deploy in the world given the timeline.”

The dramatic increase in online data theft in recent years shows that shadowy hackers routinely steal and release personal data, even though such activity is illegal. Thieves can use such information for extortion or medical identity theft. The actual re-identification of medical dossiers, however, is not a crime, although such action might constitute a breach of contract depending on the conditions set by the source of the information.

It is not hard to imagine a U.S. senator condemning a foreign country only to find his or her intimate medical data plastered on the Internet, or unscrupulous political operative leaking information about a rival candidate (the bitterness of the 2016 U.S. campaign makes such sleazy tactics easy to imagine). Rogue investors might be keen to learn inside details about the health of key corporate leaders before stock prices react to future revelations. A fanatical sports fan may want to humiliate a rival team’s star player.

“That's the key challenge: Unlike financial fraud, it's not that broad-scale sort of identification that matters, it’s the VIP identification that matters,” said Sean Nolan, former general manager of Microsoft HealthVault. “Because that's where you actually have actionable, real data that you can use.”

“The dirty not-so-secret is that data HIPAA considers anonymized isn't.”

Big pharma, big data: why drugmakers want your health records

By Ben Hirschler

LONDON (Reuters) - Drugmakers are racing to scoop up patient health records and strike deals with technology companies as big data analytics start to unlock a trove of information about how medicines perform in the real world.

Studying such real-world evidence offers manufacturers a powerful tool to prove the value of their drugs - something Roche (ROG.S) aims to leverage, for example, with last month's $2 billion (1.46 billion pounds) purchase of Flatiron Health.

Real-world evidence involves collecting data outside traditional randomised clinical trials, the current gold standard for judging medicines, and interest in the field is ballooning.

Half of the world's 1,800 clinical studies involving real-world or real-life data since 2006 have been started in the last three years, with a record 300 last year, according to a Reuters analysis of the U.S. National Institutes of Health's clinicaltrials.gov website.

Hot areas for such studies include cancer, heart disease and respiratory disorders.

Historically, it has been hard to get a handle on how drugs work in routine clinical practice but the rise of electronic medical records, databases of insurance claims, fitness wearables and even social media now offers a wealth of new data.

The ability to capture the experience of real-world patients, who represent a wider sample of society than the relatively narrow selection enrolled into traditional trials, is increasingly useful as medicine becomes more personalised.

However it also opens a new front in the debate about corporate access to personal data at a time when tech giants Apple (AAPL.O), Amazon (AMZN.O) and Google's parent Alphabet (GOOGL.O) are seeking to carve out a healthcare niche.

Some campaigners and academics worry such data will be used primarily as a commercial tool by drugmakers and may intrude upon patients' privacy.

DRUGMAKERS DELVE

Learning from the experience of millions of patients provides granularity and is especially important in a disease like cancer, where doctors want to know if there is a greater benefit from using a certain drug in patients with highly specific tumour characteristics.

In the case of the Flatiron deal, Roche is acquiring a firm working with 265 U.S. community cancer clinics and six major academic research centres, making it a leading curator of oncology evidence. Roche, which already owns 12.6 percent of Flatiron, will pay $1.9 billion for the rest.

But interest in such real-world data goes far beyond cancer.

All the world's major drug companies now have departments focused on the use of real-world data across multiple diseases and several have completed scientific studies using the information to delve into key areas addressed by their drugs.

They include diabetes studies by AstraZeneca (AZN.L) and Sanofi (SASY.PA), joint research by Pfizer (PFE.N) and Bristol-Myers Squibb (BMY.N) into stroke prevention, and a Takeda Pharmaceutical <4502.T> project in bowel disease.

"It's getting more expensive to do traditional clinical trial research, so industry is looking at ways it can achieve similar goals using routinely collected data," said Paul Taylor, a health informatics expert at University College London.

"The thing that has made all this possible is the increasing digitisation of health records."

Significantly, the world's regulators are taking notice.

U.S. Food and Drug Administration (FDA) Commissioner Scott Gottlieb - the gatekeeper to the world's biggest pharmaceutical market - believes more widespread use of real-world evidence (RWE) could cut drug development costs and help doctors make better medical choices.

Under the 21st Century Cures Act, the FDA has been directed to evaluate the expanded use of RWE. "As the breadth and reliability of RWE increases, so do opportunities for FDA to also make use of this information," Gottlieb said in a speech last September.

The European Medicines Agency, too, is studying ways to use RWE in its decision making.

WHOSE DATA IS IT ANYWAY?

But the growth of real-world evidence also raises questions about data access and patient privacy, as Britain's National Health Service (NHS) - a uniquely comprehensive source of healthcare data - has found to its cost.

An ambitious scheme to pool anonymised NHS patient data for both academic and commercial use had to be scrapped in 2016 after protests from both patients and doctors.

And last year a British hospital trust was rapped by the Information Commissioner's Office for misusing data, after it passed on personal information of around 1.6 million patients to artificial-intelligence firm Google DeepMind.

Sam Smith, a campaigner for medical data privacy at Britain's MedConfidential, is concerned drugmakers' RWE studies are just a cover for marketing. "How much of this is really for scientific discovery and how much is it about boosting profits by getting one product used instead of another?"

Some academics also worry RWE studies could be susceptible to "data dredging", where multiple analyses are conducted until one gives the hoped-for result.

AstraZeneca's head of innovative medicines Mene Pangalos, whose company has struck several deals with tech start-ups and patient groups to gather real-world data, acknowledges ensuring privacy and scientific rigour is a challenge.

"It's a real problem but I don't think it's insurmountable," he told Reuters.

"As people get more comfortable with real-world evidence studies I think it will be much more widely used. I would like to see a world where real-world data can be used to help change drug labels and be used much more aggressively than it is today."

NEXT FRONTIER

Roche Chief Executive Severin Schwan believes data is the next frontier for drugmakers and he is betting that the Swiss group's leadership in both cancer medicine and diagnostics will put it in pole position.

"There's an opportunity for us to have a strategic advantage by bringing together diagnostics and pharma with data management. This triangle is almost impossible for anybody else to copy," he said in a December interview.

Still, even Roche cannot work alone in this new world.

"You can have a big debate about whose data it is - the patient's, the government's, the insurer's - but one thing for sure is the pharmaceutical company does not own it. So there's no choice but to do partnerships," Schwan said.

With Apple's latest iPhone update including a new feature allowing users to view their medical records, Amazon teaming with Berkshire Hathaway (BRKa.N) and JPMorgan Chase (JPM.N) on a new healthcare company, and numerous start-ups flooding in, the partnering opportunities are plentiful.

"You are going to see more deals," said Susan Garfield, a partner in EY's life sciences advisory practice. "Data already has tremendous value and it is going to have increasing value in future. The question is who is going to own and capture it."

(Reporting by Ben Hirschler; Editing by Pravin Char)

Who would want this data? The drug industry, for one. Pharmaceutical companies are major buyers of these medical records—they use them to design ads to doctors and target potential patients. Other buyers include IMS Health, a provider of prescription data, also used by drug companies; OptumInsight, a division of UnitedHealth Group, the country’s biggest health insurer; and WebMD, which uses the data to tailor information found on their website.

Defining Real World Data and Real World Evidence Real World Data (RWD) are data used for decision making that are not collected in conventional randomised controlled trials (RCTs). Real World Evidence (RWE) is the evidence generated from RWD. A linked concept is ‘Big Data,’ which is generated from combing multiple sources of RWD. At the heart of RWD and RWE is patient-centricity. RWD can be clinical, economic and humanistic and come from diverse sources; from patient records to social media and nearly everything in between. However, this proliferation of data also poses challenges for methods and analysis. Further, companies need to keep tabs on RWD and RWE in their fields as, even if they are not generating this data, others might release findings that necessitate a company response. Real World Data and Real World Evidence for Market Access Interest in RWD and RWE is increasing, with various stakeholders – including companies, regulators, Health Technology Assessment (HTA) agencies, payers, providers and patients – leveraging the information to inform their decisions. This data is proving particularly relevant to some HTA agencies and payers as they focus on therapies that provide value for money. RWD and RWE are increasingly being seen as an enabler for market access, with case studies emerging on how RWD and RWE can ‘tip the balance’ from a no to a yes with HTA agencies, including the Scottish Medicines Consortium (SMC), the National Institute for Health and Care Excellence (NICE) and the French Authority for Health (HAS). However, RWD and RWE may need to be coupled with changes in price to secure those positive recommendations. Acceptance of RWD and RWE across agencies is not universal and challenges to using RWD and RWE with payers need to be overcome to increase their influence. For the companies themselves, meanwhile, RWD and RWE are core to outcomes-based contracts, which currently remain limited, but are expected to grow in the future. Further, RWD and RWE are helpful beyond initial market access and can be used to inform development plans and future business decisions. RWD and RWE can also help differentiate products from those of competitors. Companies leading on Real World Data and benchmarking The pharmaceutical industry as a whole recognises the importance of RWD and RWE, although there is significant diversity in terms of company approach. Pfizer is particularly busy – running one of the highest number of real world studies as a sole sponsor, based on analysis of data from ClinicalTrials.gov. Other big pharma companies are also gaining significant ground. For example, Astellas won the first inaugral UK Prix Galien award for RWE in 2016 based on their work on Dificlir (fidaxomin) for clostridium dificile infection in the English National Health Service (NHS). GlaxoSmithKline has been breaking new ground with the first pragmatic Phase III real world trial: the Salford Lung Study (SLS). The pay-off for this multi-million pound study in terms of market access aren’t yet known. Those companies who are seen as leaders in RWD and RWE share several common features, including:

Centralising RWD and RWE in the organisation, whilst at the same time striving to integrate this information across the business and work cross-functionally
Implementing RWD and RWE early in the clinical development process 
Investing in the infrastructure for RWD and RWE 
Gathering data from a diverse set of data sources and performing varied analysis to address the unique needs of various stakeholders 
Collaborating with others so as to provide access to RWD and expertise, as well as to add credibility to a company’s own RWE

Real World Data and collaboration

Companies need to collaborate with a variety of stakeholders to optimise RWD and RWE. Doing so will not only increase the insights from RWD and RWE, but also increase the acceptance and credibility of RWD and RWE by external stakeholders. Companies have a wide choice of collaborators, including Patient Powered Research Networks (PPRNs), providing a way to work with patients and access patient data.

The future for Real World Data and Real World Evidence

RWD and RWE are going to grow in importance, reflecting the increase in volume of data available, improvements in its quality and the ability to link disparate data sources. RWD and RWE will be in greater demand too, because of the trend for coverage and reimbursement decisions to be based on this information, as well as being used in outcomes-based contracts that are expected to increase in the future. Further, the types of data captured and the methods in which outcomes can be generated are also expected to evolve, with patient reported outcomes (PROs) expected to grow in importance. Beyond the pharmaceutical industry, a number of other stakeholders are expected to shape the future of RWD and RWE, with technology companies expected to emerge as a key influencer given their efforts to seek a greater presence in healthcare.

Pharmaceutical data mining is the business of collecting information relating to prescribers’ prescribing habits and then selling them to data mining companies, which then sell detailed report s on prescribing patterns to pharmaceutical companies. 23 Pharmaceutical companies buy this v aluable information to allow them to better target their sales force, allowing them to increase their marketing efficiency and greatly increase their profits. 24

https://36kr.com/p/5098222.html

Numerate

是一家开创性的AI公司，采用新型机器学习算法，致力于克服小分子药物发现的主要挑战。在心血管，神经退行性和癌症等疾病领域，采用传统的药物研发方法难以产生符合所有药物标准的先导化合物。因此，Numerate正在利用AI与传统的药物化学方法结合，预测候选药物成功与失败的因素。

Guido Lanza博士: Numerate的第一个最明显的区别就是我们成立很久了。在没有人关注AI的时候，我们已经建立了一家AI技术驱动的公司。我们的初创团队里有计算机科学家和新药研发人员，他们在临床和市场上都有化合物。这迫使我们在很大程度上隐藏了人工智能的部分，并以更为传统的平台公司方式开展业务，围绕以服务和研发合作为重点的合作伙伴关系。这个商业模式使我们能够在10年的时间里投资了近5000万美元建立了我们的技术平台，其中大部分是非稀释性的资金。

从科学的角度来看，我们的差异化在于转化能力。首先，我们能够使用非常小的数据集来解决新兴的生物学问题，即使这些数据不适合用深入学习的方法进行研究。二，我们的建模是基于3D配体信息，不需要化合物结构信息。这些能力使我们的机器学习算法能够解决表型驱动的药物研发难题，这种研发通常是低通量，高内涵的生物学问题。

另一个转化能力就是我们的ADME和毒性预测功能。在这方面，我们投资了1000多万美元，其中包括与美国国防部防威胁减少局（DTRA）的大型合同，以建立和验证一套系统，可能将先导物快速转化为临床候选药物。今天，我们与制药公司的许多合作都基于这一能力，其独特之处在于能够从过去的所有研发项目中学习，为未来的每一个化学设计和候选药物选择提供决策。

在最早阶段，AI面临的关键挑战是从相对较小的数据集中提取大量信息。例如，我们的平台使我们能够非常快速地将学术研究的的实验（特点是数据很少，低通量，高内涵）转化为完整的先导物优化阶段的项目。我们与Gladstone研究所合作开展了这项工作，现在开始与加州大学洛杉矶分校（UCLA）和梅奥诊所（Mayo Clinic）进行了几个项目。

AI面临的第二个挑战是整合单个项目产生的大量数据（例如组学omics数据）。在这方面，像Berg Health这样的公司能够集成大量数据来推动程序具有更多的可预测性。还有组合应用NLP（神经语言程序设计），以利用整体的生物学知识来做决策，从而能够解释结果，发现不可见的关联——例如沃森机器人和Benevolent（译者注：一家领先的英国人工智能公司，关注健康和药物开发）。

Numerate，Insilico Medicine，Berg Health和NuMedii这样的创业公司，也有GE和IBM这样的大公司

Biosymetrics
Atomwise

现在，我们可以将研究论文、实验数据、医疗专利以及临床病历统一建库，梳理挖掘需要的信息。例如寻找病人与正常人在分子层面上的差异，从而定位治疗靶点。

在麻省剑桥和多伦多均有分部的 AI 生物信息公司 BioSymetrics，就专门提供搜集整合分析临床数据的服务。

BioSymetrics 的机器学习平台 Augusta，可以处理、清洗原始数据用以后续分析。还有能够实时摘取最新研究论文信息的 AI 公司 nference，可以帮助用户获取医疗研究的最新进展，发现疾病之间的相关性。除此之外，还能为后续药物的临床实验设计作准备。

目前，医疗科研及临床信息的搜集和整合服务，代表 AI 公司有以下 5 家。

default

以开发针对炎症性肠道疾病的药物为例，第一步就是搜索出和结肠癌、溃疡性结肠炎等相关的基因，以及找出这些基因会影响哪些信号通路。

过去想要完成以上任务，需要从不同文献库查找资料，花费大量时间。现在，成立于 2012 年的荷兰 AI 制药公司 Euretos 可以根据药企的要求，迅速给出可能靶点的列表。

通过整合 200 个组学数据库，Euretos 打破了公开发表文章、试验、临床等各种数据间的壁垒。在给出靶点列表之外，Euretos 还会为候选靶点基因打分，评估其成为有效靶点的可能性大小。以溃疡性结肠炎（Ulterative Colitis）为例，UC 与候选靶点基因在 PubMed（国际公认最具权威的生物医学文献数据库）文章中共同出现的次数，和其相关的基因变体重合程度，相关疾病数，是否与肠炎有直接关系，都是候选靶点基因的打分标准。

除此之外，Euretos 还可以「学习」理解疾病的分子机制，直接评估靶点基因产生的蛋白对机体细胞和组织功能的影响，并预测功能失常后的表型、所引发的疾病。

从免疫系统、信号通路、分子立体结构等不同角度筛选靶点的 AI 公司还有：

default

https://www.huxiu.com/article/265069.html?rec=manual

深度智耀、智药科技、亿药科技云势软件

2016 年 11 月，Benevolent AI 与强生达成合作，强生把一些尚处于试验中的小分子化合物转交给了 Benevolent AI，进行新药开发。

辉瑞于 2016 年 12 月与 IBM 签署协议，利用 IBM Watson 系统协助 Pfizer 的免疫肿瘤药物研发。

2017 年 5 月，赛诺菲与 Exscientia 签订了一项潜在价值为 2.5 亿欧元的合作和许可交易，用于开发针对代谢疾病的小分子药物。

default

药物研发具有时间长、高投入、高风险的特点。其中，新药发现由靶点确认、分子设计、化合物合成、活性筛选等多个步骤组成，需要药物研发人员筛选上万个化合物，进行无数次实验去探索、证明与评估。

人工智能可以对药物结构、疾病病理生理机制、现有药物的功效、显微镜下的样本观察等等结果进行快速分析，大大提升新药发现的效率。有专家称，AI 未来有能力将临床前实验平均 6 年的研发时间缩短到数月。

default 例如，就有AI使用自然语言处理来筛选出现的文献，如化学图书馆、医学数据库和科学论文，从而得出可能的新候选药物的结论。

https://www.cn-healthcare.com/articlewm/20170714/content-1015992.html

大多数情况下，药物研发工作者会利用高通量筛选的方式无限扩大筛选对象以期邂逅目标化合物，提高药物发现的机率。由于不断试错的成本太高，越来越多的药物研发企业开始引入人工智能开发虚拟筛选技术，以取代或增强传统的高通量筛选过程。

药物研发企业通过运用人工智能药物研发系统，能在医药研发过程中减少人力、时间、物力等投入，降低药品研发成本。同时基于疾病、用药等建立数据模型，预测药品研发过程中的安全性、有效性、副作用等。此外，随着人工智能和机器学习的不断整合，药物研发企业有望在新药研发过程中显著地实现“去风险”，保守估计每年将节约大概260亿美元的研发成本。同时，还将提高全球医疗信息领域的效率，节约的成本价值超过每年280亿美元。

人工智能及机器学习可以应用在药物开发的不同环节，包括新药开发、药物有效性及安全性预测、构建新型药物分子、筛选生物标志物、研究新型组合疗法等。从全球的情况来看，人工智能辅助药物研发的公司比例相对较高，在研发周期长、投入大、失败率高等为特点的药物研发现状影像下，产业发展的需求量大，可达到千亿级的市场。

BenevolentAI

https://36kr.com/p/5130038.html

人工智能用于药物分子挖掘

位于英国伦敦的BenevolentAI成立于2013年，是一家致力于人工智能技术开发和应用的公司，是欧洲最大的AI初创公司，在全世界排名第五。这家公司建立了一种有望更快更好开发新药的人工智能技术，他们的目标是建立人们期盼已久的“制药企业2.0”，利用人工智能助力新药开发，避免代价高昂的临床试验失败。

BenevolentAI的核心技术是一个叫做JACS（Judgment Augmented Cognition System）的人工智能系统。他们认为可以通过人工智能把人、技术和生物学结合起来，集中处理全世界大量高度碎片化的信息，用以加速科学研究和发展。自2013年以来，BenevolentAI已经开发出24个候选药物，且已经有药物进入临床IIb期试验阶段。

国际制药巨头之一的强生公司已经与BenevolentAI达成合作协议，强生将一些已经进入临床阶段的试验药物连带专利一起特许给BenevolentAI，而BenevolentAI将利用人工智能系统来指导临床试验的进行和数据的收集。

总部位于伦敦的新锐公司BenevolentAI拥有自己的AI平台，汇总了来自研究论文、专利、临床试验和患者记录等来源的数据。在该平台及云计算技术基础上，能够表示出超出10亿个生物体的已知和推断关系，其中包括基因、症状、疾病、蛋白质、组织、物种和候选药物等。这个平台可以作为一个搜索引擎，让用户进行查询。例如，用户可以利用这个AI平台，来找到某种疾病相关的基因或化合物。平台上的大部分数据都没有注释，所以AI使用了自然语言处理技术来识别实体，并理解它们与其他事物的联系。“人工智能可以把所有的相关数据放在相关资料中，从而为专注于药物发现的研究人员们提供最重要的信息，”BenevolentAI首席执行官Jackie Hunter博士表示。

当BenevolentAI利用AI系统，寻找治疗肌萎缩侧索硬化症（ALS）的新方法时，系统标记了约100种具有治疗潜力的现有化合物。由此，BenevolentAI的研究人员们选择了五项来自英国Sheffield Institute of Translational Neuroscience的试验，利用来自患者的细胞进行测试。研究发现，在AI标记的这些化合物中，有四种具有良好的治疗效果，其中一种可以延缓小鼠的神经症状。

虽然行业内已经出现了许多具有明朗前景的应用，然而还有很多研究人员尚不清楚人工智能在其中发挥的作用。BenchSci是加拿大多伦多的一家初创公司，致力于为研究人员提供能够搜索抗体的机器学习工具。该公司在2月份发布的调研结果表明，在参与调研的330名药物研发人员中，有41％对人工智能并不熟悉。药物研发领域的专家认为，研究人员应该尽快掌握这方面的相关知识。

Berg Health

人工智能用于筛选生物标志物

众所周知，生物标志物是指可以标志系统、器官、组织、细胞以及亚细胞结构功能的改变或可能发生的改变的生化指标，可用于疾病诊断、判断疾病分期或者用来评价新药或新疗法在目标人群中的安全性及有效性。

Berg Health是位于美国波士顿的一家生物制药公司，成立于2006年。公司通过Interrogative Biology技术平台对患者样本进行高通量质谱分析，获得患者的基因组、蛋白组、代谢组以及线粒体功能等多方面的信息。在这过程中，可以从一个患者样本中获得上兆个数据点，将这些数据与患者的临床信息相结合，通过人工智能分析，详细描绘出患者体内生物系统个体化状态。根据这些信息，研究人员可以进一步发掘与疾病相关的生物标记物，检测手段和治疗方法。 default

Berg Health的研发管线

2016年10月，Berg Health公司与美国国防部宣布达成合作，利用人工智能技术开展新药研发。以寻找应对现有药物不起反应的侵入性乳腺癌治疗方案，将筛选多达25万个样本来寻找早期癌症的新生物学指标和生物标记。

Atomwise

人工智能用于新药有效性/安全性预测

成立于2012年的Atomwise是一家药物挖掘与人工智能结合领域的比较有代表性的初创公司，Atomwise的核心技术平台称为AtomNet，这是一种深度卷积神经网络，通过自主分析大量的药物靶点和小分子药物的结构特征，AtomNet可以学习小分子药物与靶点之间相互作用规律，并且根据学习到的规律预测小分子化合物的生物活性，从而加快药物研发进程。

这家公司通过与IBM超级计算机合作，通过分析数据库，并用深度学习神经网络分析化合物的构效关系，于药物研发早期评估新药风险。早在2015年，这家公司宣布寻找埃博拉病毒治疗方案方面有一些进展，在为时一周的时间内，从已有的药物中找到两种或许能用来抗击埃博拉病毒的药物。 default

总的来讲，Atomwise的商业模式是为制药公司、创业公司和研究机构提供候选药物预测服务。公司成立以来，已经与斯坦福大学、Scripps研究所等著名科研机构合作开展了27个药物研发项目，与默沙东也有药物研发合作项目。

https://www.cyzone.cn/article/178090.html

https://baijiahao.baidu.com/s?id=1606966140744603720&wfr=spider&for=pc

擅长提升效率的 AI，能不能在制药这件事上帮助人类？

一、从寻找靶点开始

药物研发漫长的流程，大致有 5 个必经的阶段：

1）找到靶点，即药物在人体内作用结合的位点；2）药物的设计、合成与筛选；3）临床前试验，测药物的有效性、安全性；4）临床试验；5）药品审批、上市。

找到治疗靶点最难的地方，在于理解致病机制。

在 AI 出现之前，往往需要几代科学家接力才有可能找到答案。以影片中的慢性粒白血病为例，从观察细胞、发现染色体变异，到理解治病机理、找到药物靶点，前前后后花费了近 30 年。

在麻省剑桥和多伦多均有分部的 AI 生物信息公司 BioSymetrics，就专门提供搜集整合分析临床数据的服务。

目前，医疗科研及临床信息的搜集和整合服务，代表 AI 公司有以下 5 家。 default

以开发针对炎症性肠道疾病的药物为例，第一步就是搜索出和结肠癌、溃疡性结肠炎等相关的基因，以及找出这些基因会影响哪些信号通路。

从免疫系统、信号通路、分子立体结构等不同角度筛选靶点的 AI 公司还有：

二、缩短设计、合成周期

找出靶点后，就是药物设计环节。

近年来市面上虽然出现越来越多的抗体、蛋白、核酸新药，但最主流的药物依然是实验室合成的小分子物质。因为小分子物质的相对质量足够小，可以充分和体内的靶点结合。同时稳定性好、不存在易降解的问题，药效期长，合成成本相对较低，给药途径广，还可口服。

不过小分子药物也有缺点。比如，研发过程中的随机性风险太高，试错成本大，无法预估副作用和药物毒性。此外如果药物本身携带太多不同的功能基团，即使设计出来合成也非常困难。哪怕药效再好，考虑到时间成本也会被药企放弃。

已于今年 4 月融到 1.15 亿美元的伦敦 AI 制药公司 BenevolentAI，用三个模型工具来克服小分子药物研发过程中的难题。

第一个模型用来找到最佳的药物靶点；第二个模型可以设计作用在靶点上分子，并选出其中最合适的；最后一个模型，用来找出合成这些分子的更快、更可靠的途径。

在药物合成中，如果一个药物分子里含有氨基、N 杂环、极性基团，合成会变得极其困难。针对这一挑战，BenevolentAI 高级机器学习研究员 Marwin Segler 曾在 Nature Chemistry 上发表过一篇文章，阐述其采用 AI 规划有机分子合成的路径，来提高小分子药物合成的成功率。

英国 AI 药物合成公司 Exscientia，正在尝试研发可以同时作用于两个靶点的药物。其首席化学家 Andy Bell 曾参与过蓝色小药丸的研发，也担任过英国帝国学院药物化学研究组负责人。

他们最新的研究成果，是一种有希望治愈感冒的分子药物 IMP-1088。原本 Andy Bell 的团队想找出针对疟疾寄生虫蛋白的药物分子，结果意外发现筛出的两种分子尽管针对的靶点不同，但一起用药效果会更好。于是他们重新合成了一个能同时作用于两个靶点新分子，即 IMP-1088。

感冒之所以难治，在于流感病毒不断突变。流感病毒 DNA 还胁迫人类的 NMT1 和 NMT2，为它们制造蛋白保护壳，瞒过免疫系统的侦查。IMP-1088 能够同时结合 NMT1 和 NMT2，解除病毒的挟持，在感冒初期消灭病毒。

专注药物设计合成的 AI 公司： ![Uploading 图片.png…]()

三、虚拟试验，节省临床前的现实试验时间

药物合成出来后，还要检测其有效性和安全性。

有效性，即评估药物与靶点结合的效果；安全性，则要观察药物是否会影响其他正常蛋白的功能。通常这个步骤是为了从上一环节中筛选出的「种子」药物里，挑出治疗效果最好、毒性最小的药，以便接下来倾注全力，让这个最有希望的「选手」上场，接受真实世界里各种试验的考量。

百度和腾讯都有参投的 AI 制药公司 Atomwise，基于多伦多大学计算生物团队的技术，开发了一套可以预测药物反应的 AI 工具 AtomNet。通过学习数百万实验数据和数千个蛋白质结构信息，AtomNet 可以预测小分子和靶点蛋白之间的结合反应及最终效果。根据预测的结果，AtomNet 还可以优化小分子结构，尽可能降低毒性。

虚拟预测可以帮助研究人员筛除安全性较低的药物分子，并大幅节省了物理世界里的实验操作时间。

此外，除了预测结合效果、评估毒性，还需要预测药物代谢。

去年年底拿到大笔投资的 AccutarBio，其开发的深度神经网络 ChemiNet，能精准预测药物的代谢过程，包括：吸收、分布、代谢、排泄（简称 ADME）。预测药物代谢能够帮助人们了解药物分子在机体的迁移路径，提前排除不理想的候选药物，缩短药物试验时间。

临床前预测药物反应的公司还有： ![Uploading 图片.png…]()

不过，虚拟实验的预测并非临床前试验的最后一步。

筛出药效最好毒性最小的候选药物后，还需要在培养细胞、动物上做试验。提供临床前研究服务的 Berkeley Lights 已经将细胞试验的操作、分析、筛选全面自动化，把原来耗时数月的基于细胞的药物筛选流程压缩到了数天。

四、优化临床试验

如果药物通过了临床前的细胞药检、动物药检，就进入临床环节。

通常来说，药物过不了临床 III 期试验，是件非常可惜的事情。已经投入巨额资金的药企会非常不甘心，迫切希望重新审视试验，从海量数据找出失败原因。为了满足药企这个需求，马里兰州一家 AI 数据分析公司 BullFrog 专门提供临床 III 期失败药物的试验数据分析的服务。

BullFrog 会先和药企沟通它们当前制药的项目情况和要求，列出分析所需的数据类型。签订完保密协议，药企就把数据传给 BullFrog。90 天之后，BullFrog 就可以提供一份分析报告，供药企优化 III 期试验。

除了分析 III 期药物失败原因，BullFrog 最近还和美国 LIBD（Lieber 脑发育研究所）达成研发合作，解决精神病科医生最头疼的开药问题。由于精神病的成因很复杂，不同的病人对药物的反应不一样，医生也无法确定哪种药物对患者个体的治疗效果最好，只能靠一种一种尝试。但这种不确定性可能会给患者造成无法估量的伤害。

为了降低不确定性，LIBD 向 BullFrog 的 AI 数据分析平台 bfLeap 提供了海量的精神病患者服药史数据。bfLeap 将这些数据处理分析后建模，根据病人个体状况预测效果最好的药物。

分析临床试验数据 AI 公司还有： ![Uploading 图片.png…]()

除了寻找靶点、设计药物、虚拟试验、临床数据分析之外，AI 还能帮助药研团队设计临床实验、招募筛选符合条件的被试人员，分析实验数据生成报告等等。

不过，AI 本质上只是一种工具，并不是万能「药神」。

药物研发是漫长复杂、凝聚了无数人类智慧的寻找过程，充满了不可预知的不确定性。AI 的角色，是尽可能寻找「确定」，帮助我们优化其中的各个环节。

wanghaisheng / wanghaisheng.github.io